Breakthroughs, Deadlines, and Self-Reported Progress: Contracting for Multistage Projects∗
Total Page：16
File Type：pdf, Size：1020Kb
Breakthroughs, Deadlines, and Self-Reported Progress: Contracting for Multistage Projects∗ Brett Green Curtis R. Taylor UC Berkeley (Haas) Duke University [email protected] [email protected] faculty.haas.berkeley.edu/bgreen people.duke.edu/∼crtaylor February 14, 2016 Abstract We study the optimal incentive scheme for a multistage project in which the agent privately observes intermediate progress. The optimal contract involves a soft deadline wherein the principal guarantees funding up to a certain date { if the agent reports progress at that date, then the principal gives him a relatively short hard deadline to complete the project { if progress is not reported at that date, then a probationary phase begins in which the project is randomly terminated at a constant rate until progress is reported. Self-reported progress plays a crucial (but non-stationary) role in implementation. We explore several variants of the model with implications for optimal project design. In particular, we show that the principal benefits by imposing a small cost on the agent in order to submit a progress report or by making the first stage of the project somewhat \harder" than the second. On the other hand, the principal does strictly worse by impairing the agent's ability to observe his own progress. JEL Classification: J41, L14, M55, D82 Keywords: Dynamic contract, Progress Report, Soft Deadline, Milestones, R&D ∗We are grateful to Bruno Biais, Simon Board, Alessandro Bonatti, Barney Hartman-Glaser, R. Vijay Krishna, John Morgan, Sofia Moroni, Steve Tadelis, Andy Skrzypacz, Toby Stuart, and Felipe Varas for useful comments and suggestions. We thank conference participants at the Columbia/Duke/MIT/Northwestern IO Theory Conference, ESSFM Gersenzee, the Finance Theory Group (FTG) Workshop, Stanford Institute of Theoretical Economics (SITE), WFA-CFAR Conference on Corporate Finance, and seminar participants at ASU (Carey), HEC Paris, Humboldt University, MIT, Toulouse School of Economics, University of Bonn, University of Chicago, University of Miami, and UNC Kenan-Flagler for useful feedback. Andy Schwartz and Andrew Steck provided excellent research assistance. 1 Introduction In this paper we analyze the following situation. A principal contracts with an agent to complete a project. The project requires the successful completion of two stages or breakthroughs in order to realize its benefits. The arrival rate of breakthroughs depends on the agent's hidden action. In particular, the agent can secretly divert funds for private benefit, thereby reducing the rate of a breakthrough. The second breakthrough (i.e., the date at which the project is completed) is publicly observed, but the first breakthrough is privately observed by the agent and cannot be verified by the principal. Both players are risk neutral and the agent is protected by limited liability. We are interested in how the principal should optimally design the incentive scheme and to what extent it relies on communication from the agent to the principal about whether progress has been made. Our investigation of this setting is motivated by three key ingredients often encountered in complex real-world projects involving research or implementation of novel technical designs. First, such projects are typically organized under an agency relationship because they require both a high level of technical expertise as well as substantial capital. Thus, there is generally some degree of separation between the expert individuals responsible for execution of a venture and the financial entity that provides its backing. Moreover, the preferences of the entrepreneur, contractor, or researcher (the \agent") regarding the timing, intensity and direction of investment are unlikely to be perfectly aligned with the preferences of the financier, end user, or institution (the \principal"). For example, Tirole(2006) suggests several reasons why the relationship between researchers and their funding sources \is fraught with moral hazard." Second, complex projects often require completion of multiple sequential stages before their benefits can be realized. For example, developing a new drug requires identifying specific chemical compounds or molecules in vitro (i.e., in test tubes) and then demonstrating their efficacy in pre-clinical (i.e., animal) trials. After this phase is successfully completed, clinical (i.e., human) trials begin, progressing through small scale and then large-scale. Each of these stages must be successfully completed before a drug can be submitted to the FDA for approval, and only after approval can the drug be brought to market, generating revenue for its developer and health benefits for society. In a similar vein, large-scale software engineering projects are often purported to follow the celebrated \waterfall model" of development involving six sequential phases: conception, initiation, analysis, design, construction, testing, and implementation (Royce, 1970). Indeed, in 1985 the United States Department of Defense codified six similar phases in their standards for working with software development contractors (DOD-STD-2167A). Other examples of multistage projects are ubiquitous. For 1 example, most large-scale procurement projects (e.g., construction, defense) involve numerous sequential stages. In venture capital, entrepreneurs generally progress through numerous stages (e.g., patenting, prototype development, and manufacturing) before realizing profits. Basic research, almost by definition, requires numerous successive advancements before its societal benefits are realized. Third, when and whether the agent makes progress on the project { that is, when and whether it transitions from one phase to the next { is often difficult, even impossible, for the principal to ascertain directly. This could obtain either because the principal lacks the technical proficiency to evaluate progress or because it is not possible to substantiate progress at a reasonable cost. In either case, project sponsors are often forced to rely on unverifiable progress reports made by the very individuals responsible for moving the project forward. Detecting fraudulent claims of progress in complex environments is, not surprisingly, notoriously difficult. Take, for example, the respective cases of extensive data falsification by Diederik Stapel, formerly a professor of social psychology at Tilburg University, and Anil Potti, formerly a cancer researcher at Duke University. The misconduct of both academics went undetected by their universities and funding agencies for a number of years. Similarly, in a 2006 survey of software practitioners, 86% of respondents reported having encountered false reports (Glass et al., 2008). The most common occurrences were in estimation and status reporting. \Respondents said that when lying happens, developers at the bottom level of the management hierarchy are most aware of the lying; they often know it's happening even when their management doesn't." These three common ingredients: agency, multiple stages, and intangible progress, give rise to the question at the heart of this paper. Specifically, if the principal cannot observe evolution of the project herself, can she nevertheless use unverifiable reports from the agent to monitor his progress and provide incentives for project advancement? Given the prevalence of misreporting, one might easily imagine that the answer to this question is `no.' In fact, we find that self-reported progress is an essential component of an optimally designed incentive scheme for promoting project development. To ensure the veracity of progress reports, the principal must, however, use them judiciously. To build intuition and a baseline of comparison for our main results, we start by analyzing a single-stage project in which only one breakthrough is needed to complete the project. In this case, the optimal incentive scheme can be implemented with a simple contract, which involves a single deadline T ∗ and a reward that depends only on the date at which the breakthrough arrives. If the agent realizes the breakthrough at τ ≤ T ∗, he collects the reward of R(τ). If a breakthrough is not realized by the deadline, the principal terminates the project. While the first-best policy never involves project termination, the use of a deadline plays a 2 crucial role in the provision of incentives in our (second-best) setting. In choosing the optimal deadline the principal faces a trade off; a longer deadline yields a higher probability of project completion but also requires giving the agent more rents in order to prevent shirking. Having established this single-stage benchmark, we analyze the situation of interest by adding a second stage to the project. Thus, there is a preliminary (e.g., research) stage, which must be successfully completed before moving on to the ultimate (e.g., development) stage. A breakthrough in the preliminary stage does not generate any direct benefits, is privately observed by the agent, and is unverifiable. On the other hand, project completion (i.e., when the second breakthrough has been made) is verifiable. In this setting, simple contracts are no longer optimal. If the principal uses a simple contract (with deadline T ∗) then as the deadline approaches, an agent who has not yet made a breakthrough will \run out of steam" ∗ at some point T1 < T and begin shirking. Thus, a simple contract can be improved upon. If the principal could observe the project stage directly, then she would simply fire the agent for lack of progress at T1, thereby averting the cost